Robot Mission-Planning Autopilot · Global AI Hackathon with Qwen Cloud

Forenly AI Systems' entry for the Global AI Hackathon Series with Qwen Cloud (Alibaba Cloud) — build production-ready agents on Qwen.

An embodied task-planning agent that turns a high-level goal ("navigate to shelf B3 and pick item 42") into a safe, sequenced robot action plan — perceive → plan → act → verify — running entirely on Qwen Cloud.

Community

Building this in the open — join the team chat on Discord: https://discord.gg/drrZBZBVz8

Track

Track 4 — Autopilot Agent — end-to-end business workflow automation; ambiguous inputs, external tools, HITL checkpoints; production-ready.

The robot mission-planning autopilot maps directly onto this track: it ingests an ambiguous high-level goal, resolves it into concrete sub-tasks via Qwen's advanced reasoning, orchestrates external perception and motion tools, and surfaces human-in-the-loop (HITL) checkpoints whenever a step exceeds a configurable confidence threshold.

Other tracks considered

• Track 1 — MemoryAgent — persistent memory, cross-session learning, efficient recall under limited context
• Track 2 — AI Showrunner — autonomous short-drama pipeline (script→storyboard→video) via Wan/HappyHorse (highest token allowance)
• Track 3 — Agent Society — multi-agent collaboration via task division, dialogue, negotiation; measurable gain over single-agent baseline
• Track 4 — Autopilot Agent ✅ selected
• Track 5 — EdgeAgent — Qwen-powered physical devices; edge-cloud orchestration, privacy-aware, graceful offline degradation

What we're building

A robot mission-planning autopilot powered by Qwen's flagship models on Alibaba Cloud. A mobile robot receives a high-level natural-language goal. The agent:

1. Perceives — pulls a scene description from the robot's sensors (camera feed → vision model → structured scene graph).
2. Plans — uses Qwen's advanced reasoning to decompose the goal into an ordered sequence of atomic actions with preconditions and postconditions.
3. Acts — dispatches each action to the robot's motion controller via a lightweight tool-call interface.
4. Verifies — after each action, re-queries the scene to confirm the expected state transition; on failure it re-plans or escalates to a human operator.

The entire planning loop runs on Qwen Cloud — the LLM backbone, tool-calling, and memory are hosted on Alibaba Cloud infrastructure.

⚠️ Required: backend must run on Alibaba Cloud

Submission requires proof of Alibaba Cloud deployment — a short recording (separate from the demo) plus a link to a code file in this repo that demonstrates use of Alibaba Cloud services/APIs. Plan the deployment early.

Architecture

See ARCHITECTURE.md — required diagram: how Qwen Cloud connects to backend, database, frontend.

High-level flow:

High-level goal (text)
        │
        ▼
  Qwen Cloud LLM  ←──── scene graph (perception tool)
  (planning loop)
        │
        ▼
  Action sequence  ──►  Robot motion controller
        │
        ▼
  Verify state  ──►  re-plan or HITL escalation

Key dates

Date	Event
9 Jul 2026, 22:00 GMT+1	Submission deadline

Setup / prerequisites

1. Register on Devpost.
2. Sign up for Qwen Cloud (free trial) + request free hackathon credits via the coupon form.
3. Join the Qwen Cloud Discord.
4. Clone this repo and follow docs/HACKATHON_REQUIREMENTS.md for environment setup.

Submission checklist

• Track identified — Track 4 (Autopilot Agent)
• Public open-source repo with detectable OSS license ✅ (Apache-2.0, visible in About)
• All source code, assets, run instructions
• Proof of Alibaba Cloud deployment — recording + link to code file using Alibaba Cloud services/APIs
• Architecture diagram (Qwen Cloud ↔ backend ↔ DB ↔ frontend)
• Demo video ~3 min, public on YouTube/Vimeo/Facebook Video
• Text description (features + functionality)
• Optional: blog/social post for the Blog Post Prize

Judging criteria (weighted)

• Technical Depth & Engineering — 30% (sophisticated Qwen Cloud API use; custom skills, MCP)
• Innovation & AI Creativity — 30% (architecture quality, modularity, clean code)
• Problem Value & Impact — 25% (real-world relevance, productization/OSS potential)
• Presentation & Documentation — 15%

Prizes ($45,000+)

Each of 5 tracks: $7,000 cash + $3,000 cloud credits + blog feature + swag + Ambassador opportunity. Blog Post Award $500 ×10. Top 10 Honorable Mention $500 ×10. + AI Catalyst program invite.

Status

🚧 In development. Track locked: Track 4 — Autopilot Agent. See docs/HACKATHON_REQUIREMENTS.md.

Team

Forenly AI Systems · github.com/Forenly

License

Apache License 2.0